DOCK 3.7: How It Works and How to Tweak your Setup. Trent E. Balius Shoichet Lab Group Meeting 2014/10/31

DOCK 3.7: How It Works and How to Tweak your Setup Trent E. Balius Shoichet Lab Group Meeting 2014/10/31

Outline < -0.1 DOCK History Differences between 3.6 and 3.7 Sampling method Scoring function New Input file New dir structure My work Vitamin D Receptor; Receptor desolvation Methods of evaluation docking performance > +0.1 Tweaking ones docking setups: Make a residue more polar Modify Spheres Rescore the docked poses / molecules

Outline DOCK History Differences between 3.6 and 3.7 Sampling method Scoring function New Input file New dir structure My work Vitamin D Receptor; Receptor desolvation Methods of evaluation docking performance Tweaking ones docking setups: Make a residue more polar Modify Spheres Rescore the docked poses / molecules

Kuntz Group Fortran DOCK 1 (1983) C DOCK 4 2001 Anchor & Grow DOCK 2 DOCK 3 (1990) rewrite DOCK: A History rewrite C++ DOCK 5 2006 Object oriented DOCK 3.5 DOCK 6 2008 Improve sampling (Bug fixes) Shoichet Group DOCK 3.5.54 (1997) DOCK 3.6 (2010) DOCK 3.7 (2013) Hierarchical database rewrite/ clean up DOCK 6.4 2010 DOCK 6.6 2013 Rizzo Group

How DOCK 3.6 and 3.7 works Ligand Sampling database construction Preparation, Sampling, and Scoring C O O N N C C O O N orient DOCKING score Dockable database file Scoring using a grid to speed up the calculations

DOCK 3.7 is different from, better than? 3.6 Ligand Internal Sampling DB format DB2 format There are far fewer Broken Molecules DOCK 3.6 would combine incompatible branches Ligand Orienting Bipartite graph matching this method differs between version. Minimizer was removed. Setup modifications external program changes DELPHI QNNIFT Schrodinger Reduce Epic Marvin Makefile blastermaster.py Directories restructured Now outputs mol2 files

smiles Ligand Preparation Protonate ligand generate tautomers (chemaxon) 3D Conformation (mol2 file) Charge ligand Desolavation (Amsol) Conformation Expansion (Omega) Increase sampling (turn nobles in omega) Use alternative sampling engines. Database generation

How Sampling Works (Internal) Hierarchical docking of databases DOCK 3.5.54 through DOCK 3.6 Highly compact format DOCK 3.7 format has changed. Less compacted Solves some problems with the old format/tools Connectivity is represented Lorber DM and Shoichet BK., Curr Top Med Chem. 2005;5(8):739-49.

Ligand Sampling Database Construction Rigid compound Another ring seg B segment A seg C Branch (conformer) Fully Grown Conformers

How Sampling Works (Orientational) A toy example illustrating the matching sphere orientational matching algorithm Coleman, RG et al. PLoS One. 2013; 8(10): e75992.

Sphere Generation xtal-lig.pdb Xtal ligand atoms as spheres (pdbtosph) rec.pdb Sphere generation (SPHGEN) Matching spheres Perl scripts Grid box Low dielectric spheres Modify Matching Spheres: add spheres where you would like to see rings, move existing spheres around Add (or remove) Low dielectric Spheres To regions where you would like to have less (more) screening

Current Scoring Function (DOCK 3.6 and 3.7) E score = EVDW + EES + Elig, desol VDW term is based on the AMBER united-atom force field Electrostatics term PB calculation using DELPHI or QNIFFT Binding site has low dielectric by including spheres. Ligand Desolvation desolvation grid value times by the polar and nonpolar terms in ligand file General Born approximation What s Missing? Meng, et al J. Comput. Chem. 1992, 13, 505 524 Mysinger and Shoichet J Chem Inf Model. 2010, 50(9):1561-73

Receptor preparation rec.pdb Grid box Low dielectric spheres Add (or remove) Low dielectric Spheres To regions where you would like to have less (more) screening Grid generation (vdws, CHEMGRID) Grid generation (ES, Qunift) Grid generation (Desolv, Solvmap) Protonate (Reduce) Vdw Parm Text file Modify hydrogens /protonation state Soften by changing repulsive exponent not recommended Partial Charge Text file Tart residues Modify charge file

Blastermaster.py vs Makefile Both are to prepare a pdb / protein ligand pair for docking All work done in working directory All docking files are copied to dockingfiles Blastermaster_mod.py allows one to tweak receptor file for protonation state, tarting, etc.: --addnohydrogensflag

DOCK Input File DOCK 3.7 parameter ##################################################### # NOTE: split_database_index is reserved to specify a list of files ligand_atom_file split_database_index ##################################################### # OUTPUT output_file_prefix test. ##################################################### # MATCHING match_method 2 distance_tolerance 0.05 match_goal 5000 distance_step 0.05 distance_maximum 0.5 timeout 10.0 nodes_maximum 4 nodes_minimum 4 bump_maximum 50.0 bump_rigid 50.0 ##################################################### # COLORING chemical_matching no case_sensitive no ##################################################### # SEARCH MODE atom_minimum 4 atom_maximum 100 number_save 1 molecules_maximum 100000 check_clashes do_premax do_clusters yes no no Increase for more sampling Max is 4. To go higher you have to change the code. Allowing clashes can help sampling

DOCK Input File (continued) ##################################################### # SCORING ligand_desolvation volume vdw_maximum 1.0e10 electrostatic_scale 1.0 vdw_scale 1.0 internal_scale 0.0 ##################################################### # INPUT FILES / THINGS THAT CHANGE receptor_sphere_file../dockfiles/matching_spheres.sph vdw_parameter_file../dockfiles/vdw.parms.amb.mindock delphi_nsize 81 flexible_receptor no total_receptors 1 ############## grids/data for one receptor rec_number 1 rec_group 1 rec_group_option 1 solvmap_file../dockfiles/ligand.desolv.heavy hydrogen_solvmap_file../dockfiles/ligand.desolv.hydrogen delphi_file../dockfiles/trim.electrostatics.phi chemgrid_file../dockfiles/vdw.vdw bumpmap_file../dockfiles/vdw.bmp ############## end of INDOCK Delphi_nsize is electrostatic grid size

Code Reorganization We have moved from SVN to Github repository. Directory structure is very different. Far easier to install, test and use.

Things are structured differently in Github ls -l ~/zzz.github/dock/ total 32 -rw-r--r--. 1 tbalius bks 2737 Aug 28 14:09 README.md drwxr-xr-x. 2 tbalius bks 4096 Aug 27 11:22 analysis drwxr-xr-x. 2 tbalius bks 4096 Aug 28 14:50 common drwxr-xr-x. 7 tbalius bks 4096 Sep 11 16:53 docking drwxr-xr-x. 3 tbalius bks 4096 Aug 27 11:22 install drwxr-xr-x. 9 tbalius bks 4096 Aug 28 14:09 ligand drwxr-xr-x. 19 tbalius bks 4096 Aug 27 11:22 proteins drwxr-xr-x. 3 tbalius bks 4096 Aug 27 11:23 test Work done by Teague and Ryan.

My Projects as Illustrations Water Project Incorporating GIST into the DOCK scoring functions Vitamin D Receptor Finding new ligands < -0.1 > +0.1

Evaluation Methods Pose Reproduction (cognate docking, cross docking) Enrichment calculations Look at fragment screens and see if things seem interesting Look for promising / expect interaction If not, back to the drawing board This is subjective Prospective testing of predictions

Pose Reproduction: RMSD Calculations We can correct for molecular symmetry using the Hungarian algorithm It is important that we are obtaining the correct binding mode. We can calculate RMSDs for DOCK 3.7 runs use DOCK 6.6

Use RMSD Calc. to Compare Scoring lig rec standard gist_esw_plus_eww_ref2 pdb conf 4JM6 CcP_A 2.1668 1.5725 4JMA CcP-A 1.9497 1.9497 4JMS CcP-A 1.5665 1.5665 4JMT CcP-A 0.7843 3.2321 4JMV CcP-A 0.3990 0.3990 4JMW CcP-A 1.4758 0.7236 4JMZ CcP-A 1.3876 1.3876 4JN0 CcP-A 1.0397 1.0397 4JPT CcP-A 2.6364 0.7402 4JQK CcP-A 1.4696 1.4696 4JQM CcP-A 4.0098 0.3250 4JQN CcP-A 3.0806 3.0867 Four cases, RMSDs get better One case, RMSD gets worse Seven cases, RMSDs stay the same

Use RMSD Calc. to Compare Scoring CcP conformation A Standard With GIST 4JQM ligand 4.0098 0.3250 4JMT ligand 0.7843 3.2321

Enrichments: ROC Cuves

Log-adjusted AUC: Early Enrichment Weighted More ROC curve CcP conformation A Log adjusted ROC curve 87.23 37.63

Log-adjusted AUC: Early Enrichment Weighted More AUC logauc CcP_A: 87.23 37.63 CcP_B: 84.85 32.30 CcP_C: 86.61 34.14 CcP_D: 90.66 29.73

Log-adjusted AUC: Early Enrichment Weighted More ROC curve CcP conformation D Log adjusted AUC curve 90.66 29.73

Uses Enrichments to Compare Scoring CcP Conformation A Standard Gist Etot (Esw+Eww) # Actives = 47 # Decoys = 3297 Best performing method

Uses Enrichments to Compare Scoring CcP gateless Enrichment results a for 4 different loop conformations. CcP conf A CcP conf B CcP conf C CcP conf D Standard b 35.39 34.89 36.35 26.01 Stand+Gtot c 28.04 24.75 24.7 12.19 Stand+Htot d 37.04 31.81 34.19 29.28 a Reporting the log adjusted AUC b standard is: E score = E vdw + E Es + E lig-des c standard score plus the enthalpy and entropy (G tot ). d standard score plus only the Enthalpy (H = E sw +E ww ).

Useful Scripts to Run DOCK 3.7 ls -l ~fischer/work/vdr/masterscriptsteb/ total 100 -rw-r--r--. 1 fischer bks 2158 Apr 14 2014 0001.be_balsti_py.csh -rw-r--r--. 1 fischer bks 1609 Apr 14 2014 0002.blastermaster.csh -rw-r--r--. 1 fischer bks 2664 Apr 15 2014 0002a.blastermaster_manualProt.csh -rw-r--r--. 1 fischer bks 1642 Oct 3 13:30 0002b.alignwithchimera.movespheres.csh -rw-r--r--. 1 fischer bks 3399 Apr 14 2014 0003.lig-decoy_enrichment.csh -rw-r--r--. 1 fischer bks 3430 Oct 6 16:08 0003a.lig-decoy_enrich_NotOnQueue.csh -rw-r--r--. 1 fischer bks 1039 Oct 7 10:43 0003b.addFORTRANSTOP.csh -rw-r--r--. 1 fischer bks 2255 Apr 14 2014 0004.combineScoresAndPoses.csh -rw-r--r--. 1 fischer bks 3586 Apr 16 2014 0005.AUCplot_of-lig-decoys.csh -rw-r--r--. 1 fischer bks 1593 Oct 3 14:07 0006.VS_db2.csh -rw-r--r--. 1 fischer bks 466 Apr 11 2014 0006a.VS_db2_restart.csh -rw-r--r--. 1 fischer bks 924 Oct 7 11:25 0007.VS_combineScoresAndPoses.csh -rw-r--r--. 1 fischer bks 1037 Jun 13 12:45 0007.VS_combineScoresAndPoses.qsub.csh -rw-r--r--. 1 fischer bks 920 Oct 15 16:28 0021.rescore_DOCK6_preproc_addHcrg.csh -rw-r--r--. 1 fischer bks 1106 Oct 15 15:49 0021_chimera_dock6prep.py -rw-r--r--. 1 fischer bks 4058 Oct 15 16:13 0022.rescore_DOCK6_make_grid_sph.csh -rw-r--r--. 1 fischer bks 2346 Oct 15 16:14 0023.rescore_DOCK6_hopt_ligmin_onGrid.csh -rw-r--r--. 1 fischer bks 4417 Oct 15 17:45 0024.rescore_DOCK6_FP_for3.7.csh -rw-r--r--. 1 fischer bks 1467 Oct 15 17:26 README_covalent_temp -rw-r--r--. 1 fischer bks 574 Oct 3 13:46 README_mod_sph -rw-r--r--. 1 fischer bks 328 Oct 3 14:13 README_overview -rw-r--r--. 1 fischer bks 1551 Oct 3 13:22 README_tart -rw-r--r--. 1 fischer bks 308 Oct 14 11:17 chimera.startup.com drwxr-xr-x. 2 fischer bks 28 Oct 3 14:04 for_old_cluster drwxr-xr-x. 2 fischer bks 4096 Sep 4 20:08 zzz.inputs

Tweaking your set up Spheres Place spheres were you want your rings to go Use previous screens for ideas

SPHGEN Spheres

SPHGEN Sphere Clusters

SPHGEN Cluster1 Spheres

Ligand Atoms Spheres

Matching Spheres

Idea for Where to Put Spheres

Right Side Spheres

Tweaking your set up Tarting Making residues more polar

HISTIDINE neutral delta proton [tart2] CD2 CB NE2 CG CE1 ND1 HD1 N -0.520 C 0.526 O -0.500 CA 0.219 CB 0.060 CG 0.089 CD2 0.145 CE1 0.384 ND1-0.444 NE2-0.527 H 0.248 HD1 0.320 N -0.520 C 0.526 O -0.500 CA 0.219 CB 0.060 CG 0.089 CD2 0.145 CE1 0.384 ND1-0.364 +0.08 NE2-0.687-0.16 H 0.248 HD1 0.400 +0.08 H N CA C O hnf = his negative epsilon nitrogen HISTIDINE neutral delta proton, but more polar [NE2(-0.16) --> ND1 (+0.08), HD1(+0.08)]

HISTIDINE neutral epsilon proton [tart2] CD2 HE2 N -0.520 N -0.520 NE2 ND1 C 0.526 C 0.526 O -0.500 O -0.500 CG CE1 CA 0.219 CB 0.060 CG 0.112 CD2 0.122 CE1 0.384 ND1-0.527 CA 0.219 CB 0.060 CG 0.112 CD2 0.122 CE1 0.384 ND1-0.607-0.08 CB NE2-0.444 H 0.248 HE2 0.320 NE2-0.524-0.08 H 0.248 HE2 0.480 +0.16 H N CA C O hpf = his positive epsilon hydrogen HISTIDINE neutral epsilon proton, but more polar [ND1(-0.08), NE2 (-0.08) --> HE2(+0.16)]

Original partial charges

More polar (tart2) partial charges

HISTIDINE neutral delta proton [tart3] -0.24 CB NE2 CG CD2 +0.24 HD1 ND1 CE1 N -0.520 C 0.526 O -0.500 CA 0.219 CB 0.060 CG 0.089 CD2 0.145 CE1 0.384 ND1-0.444 NE2-0.527 H 0.248 HD1 0.320 N -0.520 C 0.526 O -0.500 CA 0.219 CB 0.060 CG 0.089 CD2 0.145 CE1 0.384 ND1-0.444 +0.0 NE2-0.767-0.24 H 0.248 HD1 0.560 +0.24 H N CA C O hnf = his negative epsilon nitrogen HISTIDINE neutral delta proton, but more polar [NE2(-0.24) --> ND1 (+0.0), HD1(+0.24)]

HISTIDINE neutral epsilon proton [tart3] +0.24 CD2-0.24 HE2 N -0.520 N -0.520 NE2 ND1 C 0.526 C 0.526 O -0.500 O -0.500 CG CE1 CA 0.219 CB 0.060 CG 0.112 CD2 0.122 CE1 0.384 ND1-0.527 CA 0.219 CB 0.060 CG 0.112 CD2 0.122 CE1 0.384 ND1-0.767-0.24 CB NE2-0.444 H 0.248 HE2 0.320 NE2-0.444-0.0 H 0.248 HE2 0.560 +0.24 H N CA C O hpf = his positive epsilon hydrogen HISTIDINE neutral epsilon proton, but more polar [ND1(-0.24), NE2 (-0.0) --> HE2(+0.24)]

More polar (tart3) partial charges (symmetric)

Rescoring Is Usefule Footprint scoring (rescore with DOCK 6) Can Rescore in my DOCK version (With GIST)

Footprint scoring TYR143 SER237 HIS397 HIS305

DOCK 3.7 is new and improved! Conclusions Code is better behaved fewer broken molecules Switched to free software ( more usable for people outside of the group) Restructuring of the directories facilitates installation and usability There are several ways to tweak docking set up: Modify Spheres Make residues more polar (e.g. HIS) Rescore results

Questions? Happy Halloween!

Extra slides

+ = L i R j j i j i R j b j i j j i i R j a j i j j i i Dr q q r B B r A A E,,,,,,, 332 ( ) ( ) [ ] ( ) ( ) [ ] ( ) ( ) [ ] + L i es es i rv rv i i av av i i p G p G q p G p G B p G p G A E 8 1 8 1, 8 1,,, interp 332,, interp,, interp L L L ( ) = R l a l p l l av r A p G,, ( ) = R l b l p l l rv r B p G,, ( ) = R l l p l es Dr q p G, 332 p = (x,y,z) where e = f(p) p 1 ligand atom i http://dock.compbio.ucsf.edu/dock_6/dock6_m anual.htm#grid

Interpolation linear f ( x) ( x x1) f ( x 2 ) + ( x 2 x) f ( x1) ( x x ) 2 1 (x 1,f(x 1 )) (x,f(x)) (x 2,f(x 2 )) (x,y,f(x, y)) bilinear: Perform 3 linear Interpolations: 2 to calculate red (from cyan); and 1 to calculate green (from red) (x a,y 1,f(x a, y 1 )) (x a,y a,f(x a, y a )) (x b,y b,f(x b, y b )) (x a,y 2,f(x a, y 2 )) Trilinear: for a cube, perform 7 linear interpolations: 4 to calculate red (from the cyan); 2 to calculate green (from red); and 1 to calculate the atomic approximation (from green) http://en.wikipedia.org/wiki/trilinear_interpolation

hydrophopic Thought experiment: Simple examples of cases Where the approximation to Hydration might fail Using dock 3.5 score This is considering the desolvation of the receptor only hydrophilic Desolvated unnecessarily -- Penalty Desolvated unnecessarily -- Benefited Desolvation necessarily -- Benefited